
AI became a hammer: how to prevent every problem from looking like a nail
95% of enterprise GenAI pilots show no measurable return. 80% of AI projects fail to deliver the value they promised. Of the $30 to $40 billion spent on enterprise GenAI, only 5% of organisations see any P&L impact at all.
The problem is rarely the technology. It is that AI gets swung at every problem, whether or not it is a nail. Boards demand an AI strategy, vendors push frontier models, and committees run pilots into every function, indiscriminately. Each failed deployment erodes trust further, in management, in vendors, and in AI itself.
This white paper argues that the discipline of not using AI has become as strategically important as the discipline of using it, and gives executives a practical way to decide. It sets out three filters, combined into a single decision tree:
- Filter 1 — AI or not AI? Diagnose the root cause first. Many data problems are really process or culture problems, and no amount of AI will mop the floor while the tap is still running.
- Filter 2 — How to source it. Embedded, everyday or custom. Align the sourcing choice to the diagnosis and start with the right expectations. Only custom AI creates a moat.
- Filter 3 — Which AI. Four lanes with very different cost, accuracy and maturity: mainstream LLMs, specialised AI, right-sized small models, or wait for AI 2.0.
The paper closes with what to do on Monday morning, European deployment examples for each lane, and a full annex of sources.
By Xavier Bekaert, Mirko Vaars and Rick Hou.







